Search

The mine safety performance evaluation and its law were taken as research objects; the human factor, mechanical equipment, management system, environment and technology were taken as mine safety state indexes; accident loss and disabling incident frequency rate as mine safety performance evaluation indexes. The classification standard of safety performance was determined according to the range of influence index and evaluation index, and then the influence index value of safety performance was discretized. According to the classification standard, the classical domain and node domain of safety performance evaluation index classification were determined. Combined with the weight of each safety performance evaluation index, the correlation degree of each mine safety performance evaluation was calculated, and then the mine safety performance was evaluated, and the mine safety performance evaluation decision system was constructed. Based on the dominance-based rough set, the evaluation law of mine safety performance evaluation decision system was analyzed. To generate more reliable laws, an efficient concept lattice reduction method was introduced to reduce the attributes of decision system. The example shows that when the concept lattice is used for reduction, there are 84 concepts in the Hasse diagram, and 24 of them can be identified. The decision system has 8 conditional attributes. And 255 power sets can be obtained under 8 conditional attributes. After eliminating the defect attribute, there are 148 reducible sets, among which there are 4 optimal reducible sets with two conditional attributes. The results show that the proposed method can be used to analyze the law of mine safety performance, so as to improve mine safety work.

In recent years, large-scale data labeling has become a huge demand and extremely challenging task. More and more machine learning and deep learning methods have been proposed to generate a variety of semantic labels for data, and used to provide data labeling service as a data labeling agent. However, even todays machine learning and deep learning model may also output a wrong label. So it is necessary to optimize the quality of the collected labeling results. Generally, there are many labeling models in the data labeling market, and each label agent has a different degree of reliability. Service buyers also need to integrate many label answers from different labels agents to get the final correct label truth. In this article, we design a novel probabilistic graph based truth discovery algorithm to estimate the true label truth of the target task and the reliability of the label agent through the collected label answers. In particular, we build a novel expectation-maximization based iterative method for truth discovery to inference label truth and estimate label agent reliability. Finally, we conduct several experiments on a real-world dataset to testify the performance of our method.

In this paper, we aim to recognize materials with combined use of auditory and visual perception. To this end, we construct a new dataset named GLAudio that consists of both the geometry of the object being struck and the sound captured from either modal sound synthesis (for virtual objects) or real measurements (for real objects). Besides global geometries, our dataset also takes local geometries around different hitpoints into consideration. This local information is less explored in existing datasets. We demonstrate that local geometry has a greater impact on the sound than the global geometry and offers more cues in material recognition. To extract features from different modalities and perform proper fusion, we propose a new deep neural network GLAVNet that comprises multiple branches and a well-designed fusion module. Once trained on GLAudio, our GLAVNet provides state-of-the-art performance on material identification and supports fine-grained material categorization.

Data augmentation is one of the most significant approaches to improve the performance of object detection. Currently, data augmentation is generally used for resolving the long tail and sample imbalance problem. However, there is an imbalance of detection performance in multi-class object detection without the impact of the above-mentioned problems. In this paper, we present an offline data augmentation method based on Cannikin's Law for augmenting the categories far below the average detection accuracy. Specifically, we first apply the statistical method to find the categories with low detection performance. Secondly, a data augmentation following the cut-paste principle is constructed. The instances from the specific categories are segmented and placed into the selected pictures in the training dataset. Lastly, the augmented pictures are selected by the similarity measurement mechanism. Extensive experiments are conducted on the Pascal VOC dataset. For the FCOS detector, the mean Average Precision (mAP) is increased from 79.12% to 83.14%. For the RetinaNet detector, the mAP is boosted from 81.6% to 83.57%. Meanwhile, the detection accuracy of the specific categories is improved by 10% at least. The experimental results show the effectiveness of our data augmentation method.

Non-maximum suppression (NMS) is a significant post-process step in most object detection frameworks, the main task of which is to find the most optimal inferences. However, the low correlation between the classification confidence score and the location accuracy of bounding box severely affects the performance of object detector. Recently, several novel loss functions about the overlap rate of candidate bounding boxes have been proposed to reduce the negative impact of the low correlation problem, but there is no solution to pay attention to the influence of the background area in bounding box on NMS. In this paper, we present a refined NMS scheme based on the distribution of corner feature and the center distance of bounding box. The correlation between confidence score and location accuracy is optimized through reducing the impact of background area in bounding box. Specially, we first construct the corner feature and reorder the confidence score by its distribution. And then, a dynamically NMS threshold strategy is built by a center distance function. Extensive experiments are conducted on the Pascal VOC and the mean Average Precision (mAP) of FSSD is improved from 78.8% to 80.1%. What's more, the proposed method can be easily embedded into most of object detection frameworks without any extra training.

Electronic health records (EHRs) data plays an important role in the development of healthcare undertaking. There are many challenges in mining EHRs, such as temporality, irregularity, sparsity, bias, etc. Thus effective feature extraction and representation are key steps before any further applications. In this paper, we propose a multi-task deep representation learning method (MTDRL) with the objective of extracting the valuable clinical information from raw data and learning an effective and interpretable patient representation. Firstly, MTDRL utilizes Bidirectional Gated Recurrent Unit (BiGRU) as an encoder to learn the hidden state vectors which are used as inputs for the following networks. This encoding part can be seen as a shared network for all tasks. Secondly, patient’s in-hospital mortality prediction and sequence reconstruction are simultaneously conducted based on the encoding network. Specifically, an attention mechanism and a fully-connected layer are incorporated in prediction task and BiGRU is implemented in the other task to reconstruct the visit sequences. Finally, we apply MTDRL to real EHRs data and the experimental results demonstrate that MTDRL is capable of learning more effective patient representation and has a significant improvement in the performance of patient’s in-hospital mortality prediction. Meanwhile, the prediction results can be effectively interpreted with the attention mechanism and provide a clinically meaningful references.

Uplink MIMO is a key component of 5G NR standards for improving the data rate. Meanwhile, TDD/FDD multi-carrier deployment is essential for 5G coverage enhancement and latency reduction. Either uplink MIMO or multi-carrier operation requires two concurrent radio frequency (RF) transmitters at UE. However, NR commercial UE is capable of at most two RF transmitters for frequency below 6 GHz. Therefore, dynamic switching of UE RF transmitter between two carriers has been recently introduced in 3GPP Rel-16 to enable uplink MIMO in multi-carrier deployments. Through dynamic transmitter switching and alternative transmission on two carriers, the advantage of more uplink slots in FDD carrier and larger channel bandwidth in TDD carrier can be fully utilized. Simulation results with link adaptation validates that, when 35 microseconds is assumed to perform the switching, dynamic transmitter switching brings 18.2% to 58.1% throughput gain over TDD single carrier transmission with uplink MIMO, and 26.9% to 59.8% throughput gain over FDD and TDD transmission without uplink MIMO.

In this work, we analyzed silicon (Si) ion implanted semi-insulting gallium arsenide substrates (SI-GaAs) by using a terahertz time-domain spectroscopy (THz-TDS) system. The main goal is to determine the ion doping profile via a nondestructive approach by using THz-TDS spectroscopy. This approach will bring a new appliance to replace secondary ion mass spectroscopy (SIMS). For that, we have carried out the ion implantation experiment with an implantation energy of 200 keV at different doses. Then, we analyzed the THz-TDS experimental data of implanted SI-GaAs substrates by the Drude model to achieve depth-dependent dielectric and electrical properties at THz frequencies.

Due to the low level of college students' professional quality and low employment rate, in recent years, enhancing the extracurricular learning ability of college students has become an urgent problem to be solved in college education. Based on learning resources and its influencing factors, this article studies the selection of extracurricular learning channels for college students through Analytic hierarchy process (AHP) and Grey association analysis, aiming to improve the extracurricular learning efficiency of college students and provide references for making policy.

This design mainly studies the structure and control design of mobile platform type lower limb rehabilitation robot. Lower limb rehabilitation robot is a mechanical design and microcomputer controlled electromechanical system. The main reason for the design is to provide rehabilitation training and treatment for patients with lower limb movement inconvenience and lower limb movement disorder. This design has carried on the original design and the detailed clear explanation in the structure aspect. Some domestic and foreign research situations of this research are enumerated to illustrate the achievements and problems to be solved. The degrees of freedom of each joint of the mobile platform type lower limb rehabilitation robot are determined. The positions of each mechanism are reasonably allocated and the parameters of each part are specifically determined. The rationality of each part and mechanism design is checked through kinematics analysis and strength check.

This paper proposes a novel composite power system reliability evaluating approach based on a splitting method. The essential concept of this method is to replicate target realizations of the sampling process to increase the occurrence frequency of the rare events, thereby improving the assessment efficiency of sampling rare failure events. In the traditional Monte Carlo simulation approach, the power system samples are classified into secure and failure states. The secure states are further quantitatively evaluated by a redundant capacity index in the splitting method to enable an adaptive design of gradual failure domains. Gibbs sampling is introduced to generate states conditional on given domains. The results indicate that the proposed method can improve the efficiency of the power system reliability evaluation, especially in the case of high-reliable power systems.

Compared to conventional speech synthesis, end-to-end speech synthesis has achieved much better naturalness with more simplified system building pipeline. End-to-end framework can generate natural speech directly from characters for English. But for other languages like Chinese, recent studies have indicated that extra engineering features are still needed for model robustness and naturalness, e.g, word boundaries and prosody boundaries, which makes the front-end pipeline as complicated as the traditional approach. To maintain the naturalness of generated speech and discard language-specific expertise as much as possible, in Mandarin TTS, we introduce a novel self-attention based encoder with learnable Gaussian bias in Tacotron. We evaluate different systems with and without complex prosody information and results show that the proposed approach has the ability to generate stable and natural speech with minimum language-dependent front-end modules.

Recently, attention-based end-to-end speech synthesis has achieved superior performance compared to traditional speech synthesis models, and several approaches like global style tokens are proposed to explore the style controllability of the end-to-end model. Although the existing methods show good performance in style disentanglement and transfer, it is still unable to control the explicit emotion of generated speech. In this paper, we mainly focus on the subtle control of expressive speech synthesis, where the emotion category and strength can be easily controlled with a discrete emotional vector and a continuous simple scalar, respectively. The continuous strength controller is learned by a ranking function according to the relative attribute measured on an emotion dataset. Our method automatically learns the relationship between low-level acoustic features and high-level subtle emotion strength. Experiments show that our method can effectively improve the controllability for an expressive end-to-end model.

Although Global Style Tokens (GSTs) are a recently-proposed method to uncover expressive factors of variation in speaking style, they are a mixture of style attributes without explicitly considering the factorization of multiple-level speaking styles. In this work, we introduce a hierarchical GST architecture with residuals to Tacotron, which learns multiple-level disentangled representations to model and control different style granularities in synthesized speech. We make hierarchical evaluations conditioned on individual tokens from different GST layers. As the number of layers increases, we tend to observe a coarse to fine style decomposition. For example, the first GST layer learns a good representation of speaker IDs while finer speaking style or emotion variations can be found in higher-level layers. Meanwhile, the proposed model shows good performance of style transfer.

In this paper, the matching problem of inverse synthetic aperture radar (ISAR) image sequences and three-dimensional (3D) reconstruction problem are analyzed. Firstly, this paper proposes an improved image matching algorithm based on Scale Invariant Feature Transform (SIFT), which realizes the association of feature points of ISAR image sequences. In the new algorithm, the transformation matrix between images is calculated. Compared with the traditional SIFT matching algorithm, the problem of few matching points is resolved, and the number of associated points is increased. Then, using the matrix factorization method, the 3D position matrix is decomposed from the observation matrix to realize the 3D reconstruction of the target. Finally, the effectiveness of the association and 3D reconstruction algorithms were verified using simulated aircraft data.

With the advent of big data era, complex optimization problems with many objectives and large numbers of decision variables are constantly emerging. Traditional research about multi-objective particle swarm optimization (PSO) focuses on multi-objective optimization problems (MOPs) with small numbers of variables and less than four objectives. At present, MOPs with large numbers of variables and many objectives (greater than or equal to four) are constantly emerging. When tackling this type of MOPs, the traditional multi-objective PSO algorithms have low efficiency. Aiming at these multi-objective large-scale optimization problems (MOLSOPs) and many-objective large-scale optimization problems (MaOLSOPs), we need to explore thoroughly parallel attributes of the particle swarm, and design the novel PSO algorithms according to the characteristics of distributed parallel computation. We survey the related research on PSO: multi-objective large-scale optimization, many-objective optimization, and distributed parallelism. Based on the aforementioned three aspects, the multi-objective large-scale distributed parallel PSO and many-objective large-scale distributed parallel PSO methodologies are proposed and discussed, and the other future research trends are also illuminated.

By applying the well-known transient current model, theoretically investigating terahertz (THz) emission from gas plasma induced by multi-color femtosecond laser pulses has been developed in this work. Instead of static tunneling ionization model, we adopted Ammosov-Delone-Krainov (ADK) tunneling ionization model [1] to calculate the tunneling ionization rate. In three-color pumping schemes, the optimal power ratio tends to require a larger amount of fundamental field with the increment of total pump power. The optimal relative phases for the 2nd harmonic and 3rd harmonic field are $0.5\pi$ and $\pi$, respectively.

Compared with the conventional "front-end"–"back-end"– "vocoder" structure, based on the attention mechanism, end-to-end speech synthesis systems directly train and synthesize from text sequence to the acoustic feature sequence as a whole. Recently, a more calculation efficient end-to-end architecture named transformer, which is solely based on self-attention, was proposed to model global dependencies between the input and output sequences. However, although with many advantages, transformer lacks position information in its structure. Moreover, the weighted sum form in self-attention may disperse the attention to the whole input sequence other than focusing on the more important neighbouring positions. In order to solve the above problems, this paper introduces a hybrid self-attention structure which combines self-attention with the recurrent neural networks (RNNs). We further enhance the proposed structure with relative-position-aware biases. Mean opinion score (MOS) test results indicate that by enhancing hybrid self-attention structure with relative-position-aware biases, the proposed system achieves the best performance with only 0.11 MOS score lower than natural recording.

Current extractive summarization models utilize semantic content and non-structured features of sentences respectively to identify the sentence importance. In this paper, we present a new approach to extractive summarization by combining semantic content and non-structured features of sentences based on convolutional neural network and recurrent neural network, called CRSum. In this model, firstly, semantic content of sentences are learned by convolutional neural network, and non-structured features of sentences are learned by recurrent neural network. Secondly, we investigate whether a sentence can be used as the summary according to the above knowledge we learned. What's more, all the predictions of CRSum model can be interpreted by visualizing semantic content and non-structured features of sentences. Experimental results on LSCTC and CNN/Daily Mail corpus show that its performance is better than that of the baseline systems and surpass the state-of-the-art model in Rouge-L.

Terahertz time-domain spectroscopy was used to investigate the optical properties of oxyfluorosilicate glasses doped with neodymium (Nd3+) ions. The refractive indices and absorption coefficients of such glass samples were found to be ~3.7 and below 10 cm−1at 0.5 THz. Exhibiting high-refractive index and low loss, the Nd3+:Oxyfluorosilicate glasses are attractive for THs device applications.

We propose and demonstrate a liquid crystal (LC) based terahertz phase shifter with a thick ( $550 \ \mu \mathrm{m}$ in total) bi-Iayer LC cell structure incorporating a thin transparent polymeric intermediate layer. The present phase shifter exhibited remarkable performance improvement. The phase shift increased from 94° to 111° threshold and drive voltages decreased from 2.2 $\mathbf{V_{RMS}}$ to $\mathbf{1 V_{RMS}}$ and from $\mathbf{10 V_{RMS}}$ to $\mathbf{8.4 V_{RMM}}$ , respectively, in comparison with those for single-layer cell structure.

We demonstrate a liquid crystal based THz spatial light modulator (SLM). Indium tin oxide or gold finger type patterns were used for transparent (>70 % at 0.08 to 1.2 THz) conductor and self-cross polarizer. The phase shift and amplitude modulation achieved are greater than 90° and 30 % in broadband in THz frequency. This SLM could become a promising candidate for high speed THz imaging application via compress sensing algorithm

This paper presents the millimeter-wave (mmWave) propagation characteristics and statistical channel model parameters at 28 GHz for fifth generation (5G) wireless communication systems. The propagation characteristics of mmWave signal in the cross-corridor environments are analyzed based on the method of shooting and bouncing ray tracing/image (SBR/IM). Propagation characteristics, including path loss models, root-mean-square (RMS) delay spread, Ricean K-factor, and correlation coefficients between channel parameters are investigated. Moreover, path loss models are established for the cross-corridor channel. Results show that the path loss exponents (PLEs) vary between 1.37 and 1.75 for line-of-sight (LOS) scenarios and the PLEs vary between 2.90 and 5.09 for non-line-of-sight (NLOS) scenarios. The statistics of channel parameters are useful for 5G communication system development and deployment.

In this paper, the method of SBR/IM(shooting and Bounding Ray Tracing/Image) is used to simulate the indoor laboratory environment, which is aimed at studying the millimeter-wave propagation characteristics. The simulation data are analyzed to get the corresponding received power and root mean square(RMS) delay of different situations. The received power of different distances between transmitters and receivers are also calculated. All the results, which are processed from the data could provide a theoretical basis for setting the location of transmitter in real indoor laboratory environments.

With the wide application of radio frequency identification (RFID) technology, the possibility of the collision among readers may increase. When the number of RFID readers is large, the dimension of the RFID reader collision problem will be huge. To solve the high-dimensional RFID reader-to-reader collision problem effectively, we improve the parallel cooperative co-evolution particle swarm optimization (PCCPSO) algorithm by adopting the hybrid adaptive strategy of the inertia weight. In addition, we make parallelism implementation of the improved algorithm. Then, we use the improved algorithm to solve the RFID reader-to-reader anti-collision problem. In the experiments, we compare the improved distributed parallel particle swarm optimization (IDPPSO) algorithm with the PCCPSO algorithm, and make Wilcoxon test on the results. The experimental results demonstrate IDPPSO algorithm has better performance.

This paper presents the propagation characteristics at 60 GHz for fifth generation (5G) millimeter-wave (mmWave) wireless communication systems. The propagation characteristics of mmWave signal in the indoor radio channels are discussed based on the method of shooting and bouncing ray tracing/image (SBR/IM). Omnidirectional path loss models, received power and root-mean-square (RMS) delay spreads statistics are analyzed in terms of line-of-sight (LOS) and non-line-of-sight (NLOS) scenarios. The results show that the path loss exponent (PLE) values in LOS environments vary between 1.56 and 1.78 and the PLE value is 3.87 in NLOS scenario. The RMS delay spreads values vary between 2.1 ns and 6.6 ns. Furthermore, the normal distribution and the Gaussian distribution models best fit the received power and the RMS delay spreads at 60 GHz, respectively.

Image and video understanding enables better reconstruction of the physical world. Existing methods focus largely on geometry and visual appearance of the reconstructed scene. In this paper, we extend the frontier in image understanding and present a method to recover the material properties of cloth from a video. Previous cloth material recovery methods often require markers or complex experimental set-up to acquire physical properties, or are limited to certain types of images or videos. Our approach takes advantages of the appearance changes of the moving cloth to infer its physical properties. To extract information about the cloth, our method characterizes both the motion and the visual appearance of the cloth geometry. We apply the Convolutional Neural Network (CNN) and the Long Short Term Memory (LSTM) neural network to material recovery of cloth from videos. We also exploit simulated data to help statistical learning of mapping between the visual appearance and material type of the cloth. The effectiveness of our method is demonstrated via validation using both the simulated datasets and the real-life recorded videos.

How to improve communication effectiveness of command and control organization is one of the core issues to be addressed in the network-centric warfare. This paper is to build a mathematical modeling from three communication effectiveness metrics: communication cost, network delay and reliability. Information capacity and delay demands based on task are also needed to constrain the target. Genetic Algorithms is adopted to optimally search the communication network topology structure, link capacity and communication routing. Experimental analysis shows that the method can be expected to achieve its objectives.

Adaptability and controllability are the major advantages of statistical parametric speech synthesis (SPSS) over unit-selection synthesis. Recently, deep neural networks (DNNs) have significantly improved the performance of SPSS. However, current studies are mainly focusing on the training of speaker-dependent DNNs, which generally requires a significant amount of data from a single speaker. In this work, we perform a systematic analysis of the training of multi-speaker average voice model (AVM), which is the foundation of adaptability and controllability of a DNN-based speech synthesis system. Specifically, we employ the i-vector framework to factorise the speaker specific information, which allows a variety of speakers to share all the hidden layers. And the speaker identity vector is augmented with linguistic features in the DNN input. We systematically analyse the impact of the implementations of i-vectors and speaker normalisation.

Credit scoring has become a very important issue in the financial industry. In order to identify good or bad credit applicants, many literatures with their proposed classification methods had been published to deal with this problem. The support vector machine (SVM) of statistical learning theory was successfully applied in classification. However it still suffers from noise sensitivity originating from the fact that all the data points are treated equally. To tackle this problem, the SVM was extended into a fuzzy SVM (FSVM) by the introduction of fuzzy memberships. In this paper, a new membership calculation method is proposed, which uses a fuzzy c-means (FCM) algorithm to deal with the classification problems with outliers or noises. In the FCM-FSVM algorithm, the training dataset is divided into many clusters by using FCM, then the membership of each of training points belongs to its class can be used to be the membership of FSVM. The proposed algorithm was applied to a credit scoring classification problem, and the results verified the effectiveness of the method.